Analyzing Nursing Records in Wound Care Using a Large Language Model

Abstract

This study aimed to summarize unstructured nursing records on cancer wound management using a large language model (LLM) and assess the quality of these summaries. This retrospective descriptive study used 80 unstructured nursing records, which were generated from the documentation of specialized cancer wound care nurses. The analysis of the records consisted of four steps: 1) selecting 21 key variables based on British Columbia Cancer Agency guidelines, 2) using an LLM to summarize records according to these variables, 3) evaluating the quality of the summaries using both quantitative and qualitative assessment methods, and 4) categorizing errors in low-quality summaries. Of the 80 nursing records analyzed, the LLM achieved complete accuracy in summarizing nursing intervention variables for cancer wounds, while accurately summarizing approximately four-fifths of the nursing assessment variables. In both quantitative and qualitative evaluations of LLM-generated summaries, factual consistency demonstrated the highest quality scores. Approximately half of the low-quality summaries were reasoning errors. These findings highlight the potential of an LLM to support treatments for cancer wound patients by summarizing unstructured nursing records.